Changes to Viewshed tool between ArcGIS 9.3 and ArcGIS 10.3
The differences are a bit subtle, so I subtracted the one layer from the other to highlight the changed output.
ArcGIS 9.3 version (above)
ArcGIS 10.3 version (above)
Difference between ArcGIS 9.3 and 10.3 (blue = no difference, white = in view in 9.3)
Following Dan Patterson's advice on another thread,
Re: PathDistance Calculation discrepancy
I made a test landscape to try to identify the discrepancy in the Viewshed tool (PathDistance is next on my to-do list). It didn't help me solve the existing problem as Arc9.3 and Arc10.3 gave the same answer in the test landscape, but it added a NEW problem: in a flat landscape, I got very strange results where the point of origin was not in view! Also note that flat points to the north (or a very high point to the N) are not in view, but points to the east of the same elevation are.
Cell size is 30 m x 30 m.
I'm not sure if the pictures loaded (it just says Viewshed.tif instead of the image on my screen), but this is even more concerning than the first problem.
Any ideas why I'm getting this behavior out of the tool? I can't see any way for the origin cell to be out of view, nor for a nearby high point (curvature of the earth isn't THAT great)
Dan Patterson for the offset in the viewshed tool, do you typically just add the offset to the input raster? I did not see an option for an offset in the tool itself.
CORRECTION: with the offset correctly applied, ArcGIS 10.3 now gives a plausibly correct answer. I still have no idea what's going on in the ArcGIS 9.3 version.
With the viewshed tool, using a 6ft offset (sorry, my data are in ft for historical reasons), here is what I get for ArcGIS 10.3 for the white-labeled point 1:
Green is in view, gray is out of view
I'm using the "FLAT_EARTH" and 0.13 (default settings) for the viewshed calculations.
The results are now coming out differently in Arc9.3 (assuming I coded everything correctly), also with "FLAT_EARTH" and 0.13:
Note the weird invisible line extending out northward from the source point. Otherwise everything looks good in 9.3. No above ground level raster as that was not an option in the viewshed tool back in 9.3.
And here is the test landscape for reference:
I also tried the viewshed2 tool with a 1 m offset, and the results were improved, but the above ground raster showed one cell requiring a 12 m offset when it was the same height as the surrounding cells!
I'm also getting NoData for some of the border cells, even though I'm setting both snapraster and the extent.
The viewshed2 tool is also not ideal, as I'm hoping for backward-compatibility with earlier versions of ArcGIS 10.
With respect to the line-of-sight calculations - thanks for the amusing mental image of a person buried eyeball-deep in the ground, but we're actually interested in this for noise propagation purposes, so it's not that implausible that a noise source would be located on or near the ground.
I always use observer points as a separate file, so I can't comment on whether your approach produces similar results. I put them into the table, since they can be varied and you can control more options... see below.
The feet meter thing.... arrrrg. make sure that everything is in one coordinate system and units are all the same (aka meters 
)
I will look in more detail soon, but have a look at the below so I am sure that everything is as it should be
Configuration of inputs for viewshed and visibility analysis
Using Viewshed and Observer Points for visibility analysis—Help | ArcGIS for Desktop
the inputs are generally located in the observer point table as in... from the help topic
Controlling the visibility analysis
It is possible to limit the region of the raster inspected by specifying various items in the feature attribute dataset, such as observation point elevation values, vertical offsets, horizontal and vertical scanning angles, and scanning distances. There are nine items in total: SPOT, OFFSETA, OFFSETB, AZIMUTH1, AZIMUTH2, VERT1, VERT2, RADIUS1, and RADIUS2.
